- Maxinsights raised a $30M seed round led by Peak XV on October 22.
- It pays more than 5,000 contractors in six countries to film work for robot training.
- It says it supplied all 200,000 hours of first-person video used to train Genesis AI’s model.
Someone wearing a camera loads a dishwasher, makes a smoothie, or washes a car. Maxinsights pays more than 5,000 contractors in six countries to do exactly that, then sells the footage as training data for robots. The Santa Clara, California, company announced a $30 million seed round led by Peak XV, with Tola Capital, South Park Commons, INCE Capital, and Alumni Ventures participating.
Maxinsights says it has captured more than 2 million hours of footage, 1.5 million of them ready for model training, and serves more than 30 customers, including frontier AI labs. Eight are on recurring contracts. It also shared the round in a LinkedIn post.
“We built the whole pipeline ourselves, from the hand-tracking models to the platform that scores every operator’s footage as it arrives, because nobody was going to sell us the pieces that make messy real-world data usable. Robots get deployed into real kitchens, not staged ones, and something unexpected happens every few minutes. Edge cases are the job, and the only way to get them is to record real work in real homes. Collecting one great hour of data is easy, and holding quality across millions of hours and thousands of people is what we built the company to do,” said Yuguang (YG) Yuan, founder and CEO of Maxinsights.
Yuan previously co-founded Mega Gear Robotics, an industrial robotics company that was acquired. Before that, the release says, he was the youngest executive team leader at Eaton, running product groups with about $150 million in annual business and more than 1,500 employees.
Chief technology officer Kun Wang was an early engineer at Google X and Waymo, the release says, then co-founded QCraft, which it describes as a unicorn with more than 1,000 employees. QCraft says its driver-assistance software is in more than 1 million vehicles. The wider Maxinsights team includes engineers from Google, Microsoft, Meta, Waymo, and Hulu.
A data shortage inside a funding boom
Investors put $47.4 billion into physical AI across 521 deals in the first half of 2026, almost four times the $12 billion raised in the second half of 2025, according to Crunchbase. The category is broad. Crunchbase counts robotics, autonomous vehicles, aerospace, drones, industrial automation, and sensors, and Waymo’s $16 billion Series D alone accounted for nearly a third of the total. Humanoid robot startups raised $8.6 billion by late July, 1.8 times all of 2025, according to Dealroom data cited in a Tech Funding News ranking.
Maxinsights argues the money is arriving faster than the data to train on. Robots learn to fold a shirt or load a dishwasher only by watching people do it thousands of times, the release says, and nobody has captured that work at the scale models require.
Its answer is a five-step pipeline: collect first-person footage, check every clip, annotate it with proprietary models, curate datasets by task or scene, and deliver raw or processed data, with customer feedback shaping the next round of collection.
The company says competitors stage tasks in mock environments and ask operators to slow their hands, while it records real work at natural speed in homes, restaurants, and hotels. Rival Mecka AI says it also collects in homes, kitchens, chemistry labs, and metal fabrication facilities, so the gap may be narrower than Maxinsights suggests.
The people doing the filming wear Maxinsights’ proprietary camera hardware, which the company says gives clearer footage than shaky iPhone or GoPro video. It hires them through staffing firms in each market and manages them directly.
Quality control as the product
Real-world footage breaks most annotation tools, Maxinsights says. Hands disappear behind objects, several people’s hands enter the same frame, operators wear gloves and rings, and motion blurs. It trained its own hand-tracking models on footage it collected, then built MaxLoop, which scores every clip as it arrives.
The company says operators now get feedback the same day instead of one to two weeks later, review costs are down roughly 80%, its human review team is 90% smaller, and it processes 450,000 hours a month.
Peak XV frames that quality control as the moat. “Almost every company in this category is selling hours. Maxinsights is the only one that treated quality control as an engineering problem rather than a staffing cost, which is what lets them operate in real life,” said Rohit Agarwal, a managing director at Peak XV, the former Sequoia Capital India and Southeast Asia. “That decision is hard to copy after the fact, and it’s why the labs that tried other vendors keep coming back to them.” That is an investor’s view, and the release names no lab that left another vendor for Maxinsights.
Peak XV split from Sequoia Capital in 2023 and manages more than $10 billion in assets. It closed $1.3 billion across three India and Asia-Pacific funds in February, with deep technology among the areas it said it would expand into. Agarwal’s profile on the firm’s website lists fintech, software, direct-to-consumer, and e-commerce as his focus, so a Santa Clara robotics data seed is a US deal for a firm best known for India and Southeast Asia. Of the other investors, Tola Capital is a Seattle firm that backs enterprise software and AI-enabled companies, and INCE Capital, founded in 2019, is based in Beijing.
Customers, rivals, and the build-it-yourself option
Maxinsights names two customers. It says it supplied all 200,000 hours of first-person footage used to train GENE-26.5, the model from Genesis AI, and is the largest data provider behind Dyna-2, the model from Dyna Robotics. Genesis raised a $105 million seed round co-led by Eclipse Ventures and Khosla Ventures in July 2025, as TFN reported. The release carries no comment from Dyna, which said on August 10 that Dyna-2 was pre-trained on more than 1 million hours of human video.
“Robots need to learn from the complexity and variation of everyday life, and carefully controlled datasets alone cannot capture that. With Genesis 26.5, we focused on footage of people performing everyday tasks naturally, with much of that data provided by Maxinsights, to help our models learn from a broader range of real-world situations,” said Johnson Wang, whom the release identifies as a cofounder of Genesis AI.
Mecka AI raised a $60 million Series B led by Sequoia Capital on October 7 and claims more than $100 million in run-rate revenue, a figure that annualises recent sales. Its reported $500 million valuation is unconfirmed, as TFN noted in its Mecka coverage, which also covers XDOF, reportedly in talks for a Series B at about $1.2 billion. Maxinsights gives no revenue figure, so there is no run rate to set against Mecka’s. On the demand side, Dream Labs, founded by former Nvidia robotics researchers, raised a $90 million seed round on October 19, led by B Capital and Kindred Ventures, for a robot model that it says learns from video of people doing everyday tasks.
Some robot makers are building their own data sources. NEURA Robotics said its Series C of up to $1.4 billion will fund NEURA Gyms, training sites where humanoid robots generate real-world data, starting with a 2,300-square-metre centre at Munich Airport, TFN reported. Maxinsights says it will use its seed money to expand data collection, advance its software and hardware, and support global operations.
The question for Maxinsights is the one its release leaves open. The figures on hours, customers, and review savings come from the company, and the only two customers it names both gather data themselves. Whether labs keep buying footage from a supplier, or bring collection in-house once they know what works, will decide if a seed round this size buys a durable business or a head start.
